Short-Term Solar Radiation Forecasting with Discrete Wavelet Transform Based Boosted Machine Learning Methods

Abstract

In this article, details are given about the study of forecasting the radiation value of the next hour, which was carried out with the data obtained from the meteorological station in Afyonkarahisar. The hourly radiation data were first decomposed with the one-dimensional discrete wavelet transform (DWT) method at different levels. Ridge regression (RR) and lasso regression (LR) models were developed for each sub-signal. In order to reduce errors, the determined models were boosted using gradient boosting machines (GBM) on the residuals. The main radiation forecast signal was created with the DWT reconstruction of sub-signals' forecasts. Statistical and graphical results showed that the best radiation forecast was obtained with the proposed DWT-BLR model.